Blog Layout

Newsroom

Explore our newsroom for our weekly wreck, press releases, and trending topics.

False Assumptions (regarding AI)

February 20, 2025

Overview


Chinese AI (Artificial Intelligence) firm DeepSeek, founded in 2023, is challenging assumptions regarding AI, its development, and its applications. In short, DeepSeek is offering an AI "model" that yields better results than other models, including those from Google, OpenAI, Meta, Anthropic, and others. Based on Meta's open-source model, DeepSeek has produced its own open-source model, meaning that the model itself is free for developers to access and modify, though users still must pay for the computing costs to run the model.


Key point: DeepSeek produced its latest "r1" model with about a $5 million investment using last-generation Nvidia hardware. Compare this to the billions spent by American firms using current-generation Nvidia hardware.

The result was a 17% decline in Nvidia's share price today and a 1.8% decline for the entire stock market.

Performance

"DeepSeek says R1 is near or better than rival models in several leading benchmarks" ¹

Paradigm Shift


This development appears to mark a paradigm shift - conclusions below.

  1. The cost to advance the frontier of AI capability is much lower than expected.
  2. The latest chip designs are NOT critical for cutting-edge development.
  3. Due to (1) and (2), open-source models have a massive advantage over closed-source models because they allow independent parties to collaborate.
  4. Smaller competitors can have a massive impact, so leaders in the space will likely continue to shift.
  5. The biggest winners may be those who effectively apply AI models, not the developers.
  6. It appears customers are open to switch models.



The below quote summarizes the concerns:

“We don’t know whether this is the ‘Sputnik Moment’ for stocks, but this is certainly a wake- up call that we are not the only game in town,” said Paul Nolte at Murphy & Sylvest Wealth Management. “To put these very high valuations in the stocks thinking they have cornered the market is a huge mistake and that is being re-rated.” ²

Good News, Bad News


The success of open-source models means lower costs and better results for customers who run the models. Furthermore, if new models can be created at a fraction of the cost of older models, it is reasonable to assume development will happen more rapidly.


Conversely, DeepSeek's success suggests that the competitive moats of closed-source AI model developers are less formidable than originally expected.



While Nvidia's latest chips may see reduced demand, it may also be reasonable to ask if Nvidia will face more direct competition. Perhaps AI models can be used to help create chip designs, thereby directly threatening Nvidia.

Caveat: Bias

All AI developers have been accused of producing biased results. In early releases, Google’s Gemini produced biased results, to which its CEO responded. DeepSeek r1 appears to produce biased responses that are supportive of the Chinese Communist regime. With an open-source model, it is possible to remove existing biases or introduce new ones in subsequent releases.

Conclusion


Hopefully this installment is helpful. Our view is that it fits the model of "faster, cheaper, better."

Share by: